Pragmatics 11:4.379-400 (2001)
International Pragmatics Association
RELATIONAL CLAUSES IN ENGLISH TECHNICAL
DISCOURSE: PATTERNS OF VERB CHOICE
Arlene Harvey
Abstract
This paper reports on patterns of verb choice in identifying relational clauses (e.g. ‘X is Y, Y is X’) in English
technical manuals. While it is obvious that specific lexical verbs will feature in identifying clauses of different
functions, e.g. mean (defining), call (naming), exemplify (exemplifying), less transparent is the distribution of
these more specific verbs and the general or neutral verb be. The findings suggest that verb choice in
(technical) identifying clauses is strongly associated with the degree of equivalence constructed between the
two central nominal groups in the clause (the Token and Value). Equivalence relations are one-to-one (rather
than one-to-many) and exhaustive (rather than semantically open). Major grammatical influences on
equivalence include nominal group structure, ergativity of the clause, and the inclusion of features (e.g.
interpersonal, logical or textual) that undermine the privileging of an experientially homogeneous world-view.
The results challenge the notions that be and specific verbs are interchangeable and that be is an unmarked
choice. On the contrary, the data reveal that under certain conditions be is the more marked choice. The
results have practical implications for teachers and students of English (in particular, students of English for
Academic and/or Specific Purposes) as well as translators.
Keywords: English technical discourse, Equative clauses, Relational verbs, Definitions, Markedness
1. Introduction
The English verb be can function as an auxiliary in the verb group, specifying tense (e.g.
Security violations are increasing) or voice (e.g. The data center is being accessed
illegally).1 Alternatively, be can function as the main verb of the clause, signaling a relation
of some kind, either attributive (copular) (e.g. The user password is secure), equative (e.g.
The dataset manager is also the dataset owner) or existential (e.g. There is a problem).
Equative clauses, also referred to as identifying clauses (Halliday 1994 cf. Halliday 1967,
1968), are the main focus of this paper. Identifying clauses are a subtype of relational
clause in which one participant in the clause, expressed in a nominal group (NG), serves to
fix the identity of a second participant, also expressed in a nominal group, as in (1):
1 This research originated in a collaborative project between Sydney University and Fujitsu
Australia Ltd. I would like to express my thanks to Christian Matthiessen, Jane Simpson and Guenter
Plum for comments on earlier versions of this work. I am indebted to Martina Temmerman for translating
parts of her PhD thesis from Dutch into English for me and to the anonymous reviewer who offered
constructive and insightful comments on an earlier version of this paper.
DOI: 10.1075/prag.11.4.02har
380 Arlene Harvey
(1) The manager (participant NG) is (identifying relational verb) the person who
manages user access to data (participant NG).
Identifying clauses can be distinguished from the copular or 'attributive' type of
relational clauses in that the former contain two co-referential nominal groups that can be
reversed without any substantive change in experiential meaning. Thus:
(2) The person who manages user access to data (participant NG) is (identifying
relational verb) the manager (participant NG).
Relational clauses, both identifying and attributive, are quite common in formal
written registers of English, e.g. in technical, scientific, academic and bureaucratic texts
(Halliday & Martin 1993; Lemke 1990a; Wignell et al. 1993); in fact, the proliferation of
these clauses is one of the indexical features of formal registers.
Various kinds of relations can be indicated in the clause, such as naming,
classifying, defining, or exemplifying, and the relational verb can be either the general verb
be or more specific lexical verbs, e.g. mean, indicate, refer to (depending, of course, on the
specific sub-relation constructed). In general, the distribution of be and lexical verbs in
English has been little studied. While there have been several studies that have touched on
the distribution of be and posture verbs (such as sit, stand, lie) in other languages, most
notably Dutch (see e.g. the collection of papers in Newman, in prep; Van Oosten 1984), the
main focus of such studies has been on the semantic differences between the lexical
variants and not on the conditions under which be (or its equivalent) might be chosen.
A common assumption seems to be that, in relational clauses at least, be is more or
less interchangeable with other more specific verbs. Halliday (1994), for instance, notes
that in relational clauses (of the intensive ‘X is A', rather than the circumstantial or
possessive types): "The most typical verb is be, and X and A are nominal groups. At the
same time, many verbs other than be also occur" (1994: 120). These more specific verbs
cluster into categories that are either identifying or attributive in function (for instance,
identifying clauses contain verbs from the equative classes). Halliday’s suggestion that be is
the most typical verb in relational clauses is echoed in Martin’s (1992) argument that be is
the default, or 'unmarked' choice in identifying clauses; that is, it is the verb chosen in the
absence of a reason for selecting otherwise (Halliday 1991a). Unfortunately, while be may
be the most typical choice, it is also the most ambiguous (Davidse 1996) due to the under-
specification of both the intended sub-relation (e.g. naming, defining, classifying) and the
direction of the identification, that is, which nominal group is being used to 'fix' the identity
of the other. In fact, Halliday (1994), in his discussion of relational verbs, seems to be most
concerned with the difficulties that semantic under-specification presents for determining
the voice of the clause as either active or passive. The distinction is critical since voice
allows the appropriate participant roles to be assigned to the two nominal groups being
related (see further Section 4).
Taking Halliday’s (1967, 1968, 1994) description of identifying clauses as her
starting point, Davidse (1991, 1992a, 1996) has provided a more detailed account of the
semiotic function of identifying clauses across different registers of English (in particular,
medicine and economics). Like Halliday, she suggests two basic types of identifying clause
based on coding orientation: Decoding or encoding. In decoding clauses, the nominal
Relational clauses in English technical discourse 381
group being identified is the more familiar and concrete element (as in dictionary
definitions, where a term is provided with a gloss: ‘An X is a Y that …’). In encoding
clauses, the nominal group being identified is the less familiar and more abstract element
(as in naming definitions in which a gloss is being given a technical name: ‘A Y that … is
called X’) (Davidse 1991, 1992a, 1996; Halliday 1967, 1968, 1994). Although Davidse
observes that different subsets of lexical verbs tend to be associated with either the
decoding or encoding orientation, she offers no explanation as to why a lexical verb might
be preferred over be (and vice versa). In her study of defining relational clauses used by
students in Dutch-speaking classrooms, Temmerman (1994) moves somewhat closer to an
explanation. She describes the frequencies of distribution between zijn ('be') and alternative
lexical verbs in prototypical and non-prototypical definitions and comments that: “The
alternatives for zijn are not all mutually interchangeable”. Indeed, their appearance depends
on whether the clause is in active or passive voice.
It is the purpose of this paper to make more explicit some of the main contextual
and grammatical influences on the distribution of be and lexical verbs with a defining
function in English technical manuals. My argument is that be is not always an unmarked
choice, but rather, under certain conditions is in fact the more marked choice. The most
critical influence on verb choice appears to be the degree of co-referentiality between the
two nominal groups, that is, the extent to which they are, grammatically and semantically
speaking, 'equivalent' to one another. I have suggested elsewhere (Harvey 1999) that the
presence or mere inference of a human agent (the participant who assigns the relation) is
critical to the degree of equivalence constructed. Equivalent relations represent a
categorical, closed and exhaustive relation between the two nominal groups and are
typically expressed in fully agentive encoding clauses containing a lexical verb (in
definitions these tend to be ‘prototypical’ or formal definitions; cf. Temmerman 1994).
Non-equivalent relations represent a non-categorical, open-ended or negotiable relation and
tend to be expressed in non-agentive decoding clauses containing the verb be (these include
‘non-prototypical’ definitions such as partial definitions, ostensive definitions and relations
of synonymy; Temmerman 1994).
Before moving on to a discussion of the theoretical and descriptive framework used
in the analysis and the results of the analysis, I will briefly describe the data.
2. Data
The data for this study consist of 166 identifying clauses that appear in three in-house
technical manuals of a multinational company, and the glossary that accompanied these
manuals. While all three manuals outline various aspects of the resource protection
function operating in the company at that time (the Resource Access Control Facility or
RACF), they differ in purpose: One manual is primarily descriptive whereas the other two
are more concerned with explaining procedures for operating RACF (Plum et al. 1992).
Although the 166 identifying clauses in the data set have a variety of functions (including
defining, naming and exemplifying) the most frequent type is defining, accounting for 136
382 Arlene Harvey
of the total 166 clauses (82%).2 Defining in fact often subsumes a number of other
functions, for instance, naming, classifying, exemplifying and describing. Several of these
are included in the well-documented Aristotelian definitional structure (or ‘prototypical’
definition, Temmerman 1994).
Aristotelian/Prototypical definition: X [Name:implicit] (Definiendum) is Y [Class] that … [Descriptive characteristics] (Definiens).
Y [Class] that … [Descriptive characteristics] (Definiens) is called X [Name:explicit] (Definiendum).
Following Temmerman (1994), a distinction is made here between prototypical
defining clauses and non-prototypical clauses. Non-prototypical definitions include only
one or several of the above functions (i.e. naming, classifying or describing in isolation, or a
combination of two, but not all three, of these). This distinction is crucial in explaining the
degree of equivalence constructed, and in turn, the typical verb choice.
3. Theoretical and descriptive framework
The main theoretical and descriptive framework used in this study, systemic functional
linguistics (see e.g. Halliday 1978, 1994; Martin 1992), emphasizes the contributions of
three areas of meaning that are expressed in clause structures and text. The two primary
‘metafunctions’ are ideational (incorporating both experiential and logical meanings) and
interpersonal. Experiential meaning encompasses what is commonly thought of as the
‘content’ of a clause or text, that which is being talked about, e.g. processes in the world,
the participants (both human and non-human) involved, and the circumstances within which
the process take place. Process types give speakers the grammatical means to ‘[impose]
order on the endless variation and flow of events’ (Halliday 1994: 106), thereby making the
communication of experience manageable. Processes can be linked to one another through
logical connections such as cause-effect, contrast, time and space, condition and so forth.
At the same time, language is used to create interpersonal relationships between language
users, for example, through the negotiation of meaning, the presentation of differing
viewpoints, and so forth. Meanwhile, textual meanings (as expressed in grammatical
systems such as Theme and Given/New) allow ideational and interpersonal meanings to be
expressed as information in text (Matthiessen 1995b). Grammatical systems most relevant
to the current analysis from the three metafunctional domains are shown in Table 1.
2 This high percentage of defining clauses is not so much an indicator of the relative frequency of
definitions in the manuals than a reflection of the focus on definitions in the original study (Harvey 1996).
Relational clauses in English technical discourse 383
Table 1: Grammatical systems realizing metafunctional meanings
Ideational
Interpersonal
Textual
Experiential
Logical
Grammatic
al System
TRANSITIVITY
Y [Participant 1] is
called X [Participant
2].
ERGATIVITY
Y is called X
[Participant: Agent].
CONJUNCTION
If …, X is called
Y.
MOOD
X is called Y.
Is X Y?
MODALITY
X may/must be
called Y.
THEME
X [Theme] is
called Y.
GIVEN/NEW
X [Given] is
called Y [New].
Systemic-functional theory does not necessarily privilege ideational meaning (i.e.
the experiential ‘content’ of a clause or text) over interpersonal or textual meaning.
However, describing the clauses in my data as ‘relational’ (rather than ‘interrogative’,
‘declarative’, ‘infinitive’ and so on) suggests that the main focus of the analysis is on
experiential meaning. The two grammatical systems that encode experiential meaning at
clause level are transitivity and ergativity.
One aspect of relational experiential meaning presented here is especially crucial to
the current analysis; this is the distinction between decoding and encoding identifying
clauses (Davidse 1992a, 1996; Halliday 1967, 1968, 1988, 1994). An analysis of coding
orientation relies on what Davidse refers to as the “janus-headed” experiential structure of
the English clause, that is, the incorporation of both transitivity and ergativity structures
(Davidse 1992b; Halliday 1994: 167). According to the transitivity model of experience,
the clause contains a Process (e.g. material, mental, relational, and so forth) (realized in the
verbal group) and one or more Participants (realized in nominal groups) associated with the
particular Process type. This core configuration of Process and Participant(s) can be
elaborated through attendant Circumstances e.g. manner, cause, time, and so forth (realized
in prepositional phrases and adverbs). A description of transitivity enables relational
clauses to be differentiated from clauses expressing other types of processes (e.g. material,
mental, etc). Ergativity, on the other hand, explains the presence or absence of ‘causality’
in the clause, and in relational clauses, allows us to determine the direction of the
identification. Both transitivity and ergativity are required to ascertain the degree of
equivalence constructed in the clause, which, in turn, influences verb choice.
Despite the apparent dominance of experiential meaning in relational clauses in
technical discourse, it is in fact the interaction between all three areas of meaning –
ideational, interpersonal and textual – that accounts for the subtleties and complexities of
verb choice in my data (described in later sections).
384 Arlene Harvey
4. Transitivity – symbolic correlation and identification
The two nominal groups in identifying clauses have dual functional roles. In the first set
of functional roles, one participant (the Token) stands in a relation of symbolic correlation
to the second participant (the Value) (Halliday 1967, 1968, 1994; Davidse 1996). The
Token and Value also enter into a relation of identification, with one participant (the
Identifier) fixing the identity of the other participant (the Identified) (Halliday 1968: 190-
92; 1994).
Table 2: Semantic relations between Token and Value in the RACF data3
Function
Token
Value
Defining
TERM
GLOSS
RACF is a resource protection function which
controls user access to resources located in
a data center Naming
NAME
REFERENT
RACF center staff the resource maintenance personnel
and auditor, chief manager included
Exemplifyi
ng
EXAMPLE
SET
the authority used to reference
or update a dataset
an access authority
Classifying
MEMBER
SET
the default, generic profile,
unregistered dataset batch
protection, and protection by
the global check function
other protection in addition to
specific name protection
Describing
FORM
FUNCTION
either a user-ID or group name the first qualifier of a dataset name
HOLDER
STATUS
the one assigned to the DASD
dataset profile with the generic
name most conforming to the
dataset name
the valid access authority
3 The ‘describing’ relations in Table 2 (form/function and holder/status) are simultaneously
identifying and descriptive in function. Such clauses are in fact intermediate between identifying and
attributive (quality) (such as the access authority is valid) (Halliday 1994: 128-29; Quirk et al. 1985,
Langacker 1987; see also Harvey 1999).
Relational clauses in English technical discourse 385
The Token (Tk) and Value (Vl) are related to one another by a process of symbolic
representation. One participant, which is abstract/general, is coded as the realization of the
other, more concrete/specific, participant (Davidse 1992a: 111). The main semantic
relations that link the Token and Value (and that occur in my data) are shown in Table 2
The Token typically (but not necessarily) conflates with the more concrete
participant in the clause. For instance, in (3) below the Token simple name is a technical
term that encapsulates the more general concept expressed in the gloss, while in (4) the
Token dataset password protection function functions as an example/member of the more
general set expressed in the Value.
(3) A simple name (Tk) is a dataset name consisting of a set of up to eight alpha-
numeric characters, beginning with an alphabetic character (Vl). (158)
(4) One of the dataset protection methods (Vl) is the dataset password protection
function (Tk). (55)
The fact that the Token and Value are (more or less) equivalent to one another is
reflected in the potential of the clause to be in either active or passive voice. Although the
verb be does not itself show voice, a specific lexical verb can reveal it.
(5) A simple name (Tk) means/refers to a dataset name consisting of a set of up to eight
alpha-numeric characters, beginning with an alphabetic character (Vl). (Active)
(158c)4
(6) One of the dataset protection methods (Vl) is exemplified by the dataset password
protection function (Tk). (Passive) (55c)
At the same time that the identifying clause sets up a coding relation between the
Token and Value, the clause constructs a kind of ‘equation’ between these two participants.
In this equation, one participant can be interpreted as a constant – that which is being
Identified, referred to as the ‘Identified’ (Id) – and the other a variable, the ‘Identifier’ (Ir).
This variable, open-ended element can be probed with a wh-question: ‘What/which one is
X?’ (Halliday 1967: 67; 1994: 123):
(7) A simple name (Id/Tk) is a dataset name consisting of a set of up to eight alpha-
numeric characters, beginning with an alphabetic character (Ir/Vl). (158) (Probe: “what is a simple name?” NOT “what is a dataset name consisting of a set of up to eight
alpha numeric characters …?”)
(8) The valid access authority (Id/Vl) is the one assigned to the DASD dataset profile
with the generic name most conforming to the dataset name (Ir/Tk). (211) (Probe: “what is the valid authority access authority? NOT “what is the one assigned to the DASD
dataset profile …?”)
4 Some examples have been modified in order to illustrate a point; these semi-
constructed examples include a ‘c’, e.g. (158c).
386 Arlene Harvey
Identifying clauses can be differentiated along two dimensions: the directionality of
the coding and the markedness of the function structure.
4.1. Coding orientation
The directionality of the coding in identifying clauses leads to two coding orientations:
decoding or encoding. In decoding clauses, the direction of the identification is from a
concrete Token as the element being identified to an abstract Value as that which serves to
identify it (Identifier). In encoding clauses, the move is in the opposite direction, that is,
from an abstract Value as the element being identified to a concrete Token as Identifier.
Further, the conflation of the different functional elements in the clause can be either
unmarked or marked. In their unmarked, or typical, realization, the functions of Identified
and Identifier conflate with the textual elements of Given and New respectively, as in (9)
below.
(9) A simple name (Tk/Id/Given) a dataset name consisting of a set of up to eight
alpha-numeric characters, beginning with an alphabetic character (Vl/Ir/New). (158)
The Identified therefore conflates with Given in its unmarked position as Theme of
the clause (that is, in clause-initial position) while the Identifier conflates with New in its
unmarked position at the end of the clause.5
The two sets of participants (Token and Value, Identified and Identifier) yield eight
clause types (Halliday 1967: 67-69, 227; 1968: 190-92; Davidse 1992a, 1996). Since four
of these assume an abstract Token and a concrete Value (not found in my data), the simpler
four-cell model (Table 3) is included here.
Table 3: The four-cell model of coding orientations
Decoding
Encoding
Unmarked
Tk/Id ^ V/Ir
Vl/Id ^ Tk/Ir
Marked
Vl/Ir ^ Tk/Id Tk/Ir ^ Vl/Id
Note: ^ indicates order of constitutents
5 New information can be interpreted as either information that has not been previously mentioned,
information that is not recoverable from the context, and/or information that is newsworthy (Halliday 1994:
59, 298; Fries 1994: 230-33). Because the Identified and Identifier functions often conflate with Given and
New, it is tempting to consider these functions as expressing similar meaning. However, Identified and
Identifier are ideational and determine 'identity', while Given and New are textual and are concerned with
information status (Halliday 1994: 124).
Relational clauses in English technical discourse 387
Examples of these four conflations, along with the probe clarifying the direction of the
identification, are provided in Table 4 (New information is in bold).
Table 4: Identifying clauses with unmarked and marked coding orientations
Decoding
Encoding
Unmarked
Cautions on Attributes: A group-user
attribute (Tk/Id) is a user attribute
that applies when a user is logged on
with a specific group to use the data
center (Vl/Ir). (92)
(Probe: “What is a group user
attribute?”)
This security method is called default
value protection. Other security methods
RACF offers (Vl/Id) are batch protection
for unregistered datasets, generic name
protection, and full resource name
protection (Tk/Ir). (27)
(Probe: “What are other security methods
RACF offers?”)
Marked
Resource: That requiring protection
by RACF (Vl/Ir) is called a resource
(Tk/Id). (150)
(Probe: “What is a resource?”)
Each data center user is identified by a
user-ID. A user-ID or user-ID card is
used to validate a legitimate user of the
data center. Only legitimate users can use
the data center. The user-password
(Tk/Ir) is an important method of
validating a user (Vl/Id). (205)
(Probe: “What is an important method of
validating a user?”)
Although identifying clauses in general are reversible (as suggested earlier),
there are in fact differences in the voice potential between the decoding and encoding
subtypes (Halliday 1994: 165): encoding clauses can be easily made passive whereas
decoding clauses display a more ‘strained’ passive (Davidse 1992a). This difference is
made clearer when a lexical verb is substituted for be. Thus, while the encoding passive
structure in (10’) below is quite acceptable, the decoding passive structure in (11’) is
clearly awkward and requires grammatical restructuring (what is meant by) to make it
more acceptable.
(10) Batch protection for unregistered datasets, generic name protection, and full
resource name protection (Tk/Ir) exemplify other security methods RACF offers
(Vl/Id). (Encoding: active) (27c)
(10’) Other security methods RACF offers (Vl/Id) are exemplified by batch protection
for unregistered datasets, generic name protection, and full resource name
protection (Tk/Ir). (Encoding: passive) (27c)
(11) A group-user attribute (Tk/Id) means a user attribute that applies when a user is
logged on with a specific group to use the data center (Vl/Ir). (Decoding: active)
(92c)
388 Arlene Harvey
(11’) ?A user attribute that applies when a user is logged on with a specific group to
use the data center (Vl/Ir) is [what is] meant by/defined by group-user attribute
(Tk/Id). (Decoding: strained passive) (92c)
Differences in voice potential in encoding and decoding clauses can be attributed to
differences in the ergativity structure of the two clause types: Decoding clauses are middle
(non-agentive) while encoding clauses are effective (agentive).
4.2. Ergativity
Unlike the transitivity system, which is a process + extension model, the ergative system is
an ‘instigation of process’ model (Davidse 1992b: 109). In encoding identifying clauses,
the Token is analyzed as an Agent because it functions as an 'embodiment' or (relatively)
concrete expression of the Value (Davidse 1996: 390). In other words, the Token ‘causes’
the Value to have concrete expression. Thus, in (12) below, the Token/Agent offers several
specific, concrete examples of the more general set of security methods offered by RACF.
(12) Other security methods RACF offers (Vl/Medium) are batch protection for
unregistered datasets, generic name protection, and full resource name protection
(Tk/Agent). (Encoding: effective) (27) 6
Some encoding clauses are doubly effective in the sense that they allow a secondary
Agent, the instigator of the process (called the ‘Assigner’ in relational identifying clauses)
(Halliday 1994), to effect (or assign) the relation between the Token and Value. In (13), for
example, the Token RACF centre staff is agentive in the sense that it enables the
embodiment of the abstract meaning/Value, while the secondary Agent (Assigner) ‘X’ is
responsible for assigning this Token to the Value. 7
(13) The resource maintenance personnel and auditor, chief manager included
(Vl/Medium) are called RACF center staff (Tk/Agent 1) [by X (Assigner/Agent 2)].
(Encoding: effective, double agency) (132c)
The effective ergative structure of encoding clauses reflects the ‘tight-fit’ between
the Token and Value (i.e. X = Y; Y = X) and the relation can be interpreted as categorical
and exhaustive.
6 Just as the Actor is the participant most closely associated with the (material) Process in the
transitive model (with the potential for a Goal to be added), it is the Medium that is most closely associated
with the Process in the ergative model. The Medium is the “key figure in [the] process; … the one through
which the process is actualized, and without which there would be no process at all.” (Halliday 1994: 163) The configuration of Medium + Agent can be expanded to include the instigator of the process.
7 This example has been partially constructed because there are no instances in my data of an
explicit Assigner (although an Assigner can be inferred from the passive construction).
Relational clauses in English technical discourse 389
In contrast, middle decoding clauses, as in (14) below, contain neither a primary nor
a secondary Agent: The Token is not an embodiment of the Value, and there is no
possibility of a secondary Agent instigating the relation. That is, 'Y is meant by X by A' is
not possible:
(14) A group-user attribute (Tk/Medium) means a user attribute that applies when a user
is logged on with a specific group to use the data center (Vl/Range). (Decoding:
middle) (92c) 8
The middle ergative structure of decoding clauses suggests a ‘loose fit’ – the
relation is more open-ended than the relation constructed in encoding clauses. In other
words, decoding clauses, in general, construct a less ‘equivalent’ relation than do encoding
clauses (Harvey 1999).
5. Equivalence as experiential deautomatization
The multifunctional structure of the most equivalent of relational clauses – the prototypical
definition – reflects, at a micro level, the macro processes operating in technical discourse
(Harvey 1999). In technical discourse, the experiential sub-component of the ideational
metafunction is mostly foregrounded or ‘deautomatized’ (to use Mukarovsky’s 1977 term)
in the structure ‘X is Y’. At the same time, logical, interpersonal and textual meanings are
typically backgrounded or ‘automatized’. Following Mukarovsky, Halliday (1982: 130)
describes the process of automatization as one in which “words and structures [function]
simply as the most neutral, or unmarked, expression of the meanings that lie behind them”
(see also Givón 1995; Jakobson 1932/1984). To put this another way, automatized
grammatical structures “[realize] the semantic selections in an unmarked way – getting on
with expressing the meanings, without parading themselves in patterns of their own”
(Halliday 1982: 135). Experiential deautomatization represents a kind of ‘ideational
privileging’ (Lemke 1990b) which in technical and technocratic discourses constructs an
apparently homogeneous world-view, the validity of which is not open to questioning or
scrutiny.9 This impenetrable world-view is most clearly reflected in relations of
equivalence, in which ‘X is Y’ and ‘Y is X’.
8 The Range is a ‘quasi-participant’ that “restates the process or specifies its extent or scope”
(Halliday 1994: 149, 167; Davidse 1992b: 125)
9 In formal definitions, a depersonalizing homogeneity is evidenced in the backgrounding of the
source of the definition, ‘X = Y’ cf. ‘We call X Y’. However, the encoding defining/naming clause in the
passive voice (‘Y is called X’) is only partially personalized because the human Assigner remains implicit;
at the same time, the clause is more effective than its decoding counterpart, precisely due to its potential for
an Assigner. The inferred Assigner thus lends the weight of authority to the specified relation, but at the
same time does not overtly ‘personalize’ the text (Harvey 1999).
390 Arlene Harvey
5.1. The grammatical construction of equivalence and non-equivalence
Equivalent relations possess a number of distinguishing features. First, the clause must
include (at least) two nominal groups, both of which must contain a logical head (noun).
For instance, (15) below contains two nominal groups representing 'things' that can be
equated. This prototypical (or 'complete') definition includes a definiendum (the Token),
and a definiens (the Value), with the latter containing both a class word and defining
characteristics.
(15) Group (Tk/Id) (definiendum) is a user-group (class) that is formally registered in the
RACF management dataset (defining characteristics) (Vl/Ir) (definiens). (Decoding)
(84)
On the other hand, the nominal group in the Value in (16) below does not contain a
logical head (class) and is thus a non-prototypical (or ‘partial’) definition; the Value is
constructed as a process rather than as a thing that can be equated to another thing.
(16) Authority revoked (Tk/Id) (definiendum) means [0 class] that a user cannot use the
data center (defining characteristics) (Vl/Ir) (definiens). (Decoding) (20)
Even if the clause does contain two nominal groups that can be equated, and both
contain a logical head, the relation need not be equivalent. To express equivalence, the
relation between the Token and Value nominal groups must also be exhaustive (rather than
open), one-to-one (as opposed to one-to-many), and presented as universally valid and non-
negotiable. Whereas (complete) prototypical definitions in the default case express a one-
to-one and exhaustive relation, other functions appear to be inherently non-exhaustive (e.g.
describing relations) while yet others seem to be intrinsically one-to-many (e.g.
exemplifying relations).10
Identifying relations incorporating descriptive aspects typically express non-
equivalence since the quality/attribute is often non-intrinsic, transient or situationally
specific (i.e. not universal). For instance, in (17) below (a holder-status relation) it is
possible to imagine a situation in which a new user-password will violate the rule specified
in the Token; in (18) (a form-function relation), the first qualifier need not be a group name,
although it usually is.
10
Compare this with the following expanded definition (i.e. extending over more than one
sentence), the final clause of which contains two non-coreferential nominal groups. These Tokens are both
associated with their own Values and thus construct two one-to-one relations rather than a one-to-many
relation (i.e. the Values are the two operations referred to anaphorically): The attribute of authority revoked
(REVOKE) is designed to temporarily prevent a user from using the data center. The attribute of authority
resumed (RESUME) restores the authority of a user whose authority has been revoked. A user in this
position can no longer use the data center (once the current job or TSS session is completed). The date can
also be specified to revoke or resume the user’s authority from a particular date. These operations (Vl/Id)
are referred to as authority auto-revoked with a date specification and authority auto-resumed with a date
specification (Tk/Ir). (Encoding) (17)
Relational clauses in English technical discourse 391
(17) Sometimes a current user-password must be changed because it has become known
to someone or for some other reason. Also, users who have just been registered can
only use the data center if they specify a new password during their first initial
logon. A new user-password (Vl/Id) must be a character string that has not been
used (or “specified”) for some generations (Tk/Ir). (Encoding) (108)
(18) The first qualifier of the VSAM catalog dataset name (Vl/Id) is usually a group
name (Tk/Ir). (Encoding) (89)
Notice that both of these clauses contain interpersonally deautomatizing features
(the modal verb must and modal adjunct usually respectively), whose inclusion has the
effect of making the validity or necessity of the relation negotiable (Flowerdew 1991).
Exemplifying clauses, on the other hand, are essentially non-equivalent since the
example offered is assumed to be a representative of a larger set. One example may be
offered in the clause, either implicitly, as in (19) below (in which we assume there are other
access authorities) or explicitly, as in (20) (in which the example cited is presented as just
one of the dataset protection methods on offer). Alternatively, several examples may be
provided, as in (21). In each case, though, the examples do not exhaust the full set. 11
(19) The access authority is an authority that enables a user to access resources. For
example, the authority used to reference or update a dataset (Tk/Ir) is regarded as an
access authority (Vl/Id). (Encoding) (5)
(20) For each dataset, the dataset manager registers the users whose use of the dataset is
authorized. Only those users whose access to the data center is authorized can
access the approved datasets. One of the dataset protection methods (Vl/Id) is the
dataset password protection function (Tk/Ir). (Encoding) (55)
(21) Data center users have a variety of objectives such as to calculate wages, and to
calculate sales. Users are grouped according to their objective. Groups (Vl/Id) may
be divisions or departments, laboratories, or project teams (Tk/Ir). (Encoding) (94)
The unmarked equivalence status of the relational functions (e.g. defining,
exemplifying, and so forth) can be disrupted. Functions that are typically presented as one-
to-one can be construed as one-to-many, for instance. The disruption of our expectations
regarding the typical equivalence status of a function is usually signaled by
deautomatization of one or more of the non-experiential types of meaning: Logical,
interpersonal and textual.
11
Although it is unusual for all examples of a set to be included, in identifying clauses constructing a member-set relation, all members can be provided. For instance: RACF provides other protection in addition to specific name protection. These functions (Vl/Id) are the default, generic profile, unregistered dataset batch protection, and protection by the global check function (Tk/Ir). (Encoding) (192) Thus, a relation can be one-to-many but non-exhaustive (as in example-set relations) or one-to-many and exhaustive (as in member-set relations).
Logical deautomatization is achieved in two main ways. There may be a dependent
392 Arlene Harvey
clause or prepositional phrase expressing a condition or specific situation. This has the
effect of reducing the universality of the relation by suggesting conditions under which the
relation holds or through the specification of situations within which the relation is valid.
a) If [Condition] is the case, X = Y (or Y = X)
b) In the case of [Condition], X = Y (or Y = X)
(22) When the RACF data set is duplicated, one of them (Vl/Id) is called the main
management data set (Tk/Ir), and the other (Vl/Id) is called the sub-management
data set (Tk/Ir). (Encoding) (103)
Alternatively, a one-to-many (rather than one-to-one) relation may be constructed,
with the Token constituted by more than one element, or alternatively, the Value constituted
by more than one element. That is, rather than one of the two nominal groups lacking a
logical head (as suggested earlier) (and thus two things that can effectively be equated), one
of the nominal groups contains two or more (typically non-coreferential) heads:
c) Token (A, B …) = Value
d) Value (A, B …) = Token
The following definition exhibits both types of logical deautomatization: Specific
conditions under which the relation holds (expressed in the dependent if and when clauses)
along with logical extension at nominal group level (i.e. two non-coreferential heads within
the Token). 12
(23) The current group (Vl/Id) is the group name specified when the user is used, and
the default group if the group name is not specified (Tk/Ir). (Encoding (37)
In summary, some relational functions can be considered to be default one-to-one
(e.g. defining and naming) while others can be regarded as default one-to-many (e.g.
exemplifying, describing, and so on). Functions such as definitions, which are typically
expressed as equivalent (in the sense of being both one-to-one and exhaustive) can, in
marked cases, be non-equivalent (i.e. either non-exhaustive or one-to-many). However,
non-equivalence in such cases tends to be accompanied by the deautomatization of non-
experiential features as a way of signaling the disruption.
5.2. Relational verbs as reflection of equivalence
12
Compare this with clauses in which there are two coreferential heads, for example:
Users who administer RACF operations and those who manage groups and resources (Vl) are called managers. (Tk).
My claim is that the degree of equivalence that is grammatically constructed between the
two (main) nominal groups in identifying clauses is the main motivator of verb choice, at
Relational clauses in English technical discourse 393
least in my data set. Two types of verbs are distinguished here: The general/neutral verb be
and specific lexical verbs. Be (along with graphological relators such as the comma, used
parenthetically, and the colon) signals neither the specific sub-relation intended (e.g.
defining, naming, exemplifying) nor voice; specific verbs (e.g. call, know as, refer to as,
mean, explain, regard as, indicate, refer to), in contrast, are unambiguous in these respects.
The first main finding of the analysis (Table 5) is that decoding clauses usually contain the
verb be (or a graphological relator) (84%) while encoding clauses typically contain a
specific verb (72%).
Table 5: Distribution of verbs (relators) and coding orientation
Decoding
Encoding
General
(Be/Graphic)
70 (84%)
23 (28%)
Specific
13 (16%)
60 (72%)
Total
83 (100%)
83 (100%)
These combinations may be the result of semantic compatibility between coding
orientation and verb type (Harvey 1999). For instance, the general and non-exhaustive verb
be (and the graphological relators) complements the semantic open-endedness and lack of
exhaustiveness inherent in (middle) decoding clauses. In contrast, specific verbs appear to
complement the closure and exhaustiveness of (effective) encoding clauses. Moreover, the
concept of semantic compatibility can be related to the notion of equivalence to make the
more general observation that equivalent relations prefer specific verbs, while less
equivalent relations prefer be (or relator). This explains not only the major coding
orientation/verb combinations mentioned above, but also accounts for the 16% of decoding
clauses that take a specific verb and (all but one instance of) the 28% of decoding clauses
containing be.13
The results suggest that decoding clauses (Table 6) construct non-equivalent
relations in 100% of the total 83 decoding clauses, regardless of whether the verb is neutral
or specific. This pattern holds for functions that, by default, express one-to-one relations
(e.g. definitions) as well as those that, by default, express one-to-many relations (e.g.
describing and exemplifying). In particular, the majority of decoding clauses in the data
set are defining (but not defining/naming) and express one-to-one relations that are open
(rather than exhaustive) (94% of the total 80 instances). The defining clause may be partial
(usually missing a class word in the Value) (24) or complete but semantically open-ended
as reflected in its middle ergativity (25).
(24) The revoke attribute (REVOKE) (Tk/Id) indicates that a user cannot access the data
center (Vl/Ir). (Decoding: middle/partial) (156)
13
For discussion of grammatical probabilities from a systemic-functional perspective, see Halliday (1991b), Halliday & James (1993), Nesbitt & Plum (1988), and Plum & Cowling (1987).
394 Arlene Harvey
(25) Library (Tk/Id) is the DASD dataset holding the program specified by the program
path control function (Vl/Ir). (Decoding: middle/complete) (97)
Table 6: Decoding clauses
The results for encoding clauses (Table 7) suggest more variability depending on the
function of the relation. Unlike decoding clauses, encoding clauses that construct a one-to-
one relation tend to be exhaustive rather than open (92% of the total 83 encoding clauses).
The most common pattern for one-to-one encoding clauses is that they contain a specific
(passive) defining/naming or naming verb (93% of the total 60 one-to-one clauses). The
most common pattern for one-to-many encoding clauses is that they consist of functions
other than defining (which is unsurprising) and contain be (91% of the total 23 one-to-many
clauses).
The majority of clauses in the data therefore conform to the pattern of equivalent
relations combined with specific verbs, and non-equivalent relations combined with the
neutral verb/relators. However, there are six clauses – two decoding and four encoding
(indicated in Tables 6 and 7 in bold-type) – that require discussion. Although five of these
clauses can be explained with reference to the general principle, they illustrate how default
patterns can be disrupted, and how such disruption is signaled. (The sixth clause, asterisked
in Table 7, does not conform to the pattern and is discussed further below.)
First, relational clauses with a combined defining/naming function are mostly
expressed in encoding clauses and grammatically construct the highest degree of
equivalence. However, there are two decoding clauses in Table 6 that are defining/naming
Relational clauses in English technical discourse 395
in function, contain a specific verb, but are non-equivalent. According to the general
principle, this seems to be a semantically incompatible combination. The first of these two
decoding clauses appears below in clause (d) of (26):
Table 7: Encoding clauses
(26) Standard DASD dataset naming rules
a) An alphanumeric character string consisting of a maximum of eight characters
starting with an alphabetic character is known as a simple name.
b) A character string linking two or more simple names with a period “.” is called a
qualified name.
c) If a dataset name consists of a simple name, the simple name itself (Vl/Id) is
called a first qualifier (Tk/Ir). (Encoding)
c) If the dataset consists of a qualified name, the first simple name in the dataset
name (Vl/Ir) is called a first qualifier (Tk/Id). (Decoding) (66-67)
(Probe for clause (c): “what is the simple name itself [called]?” Probe for clause (d):
“in this case what is called the first qualifier”)
The defining clauses in (c) and (d) together set up a non-equivalent relation. Clause
(c) is a typical encoding defining/naming clause with an unmarked Id ^ Ir structure. The
element being sought in this definition is the name that is given to the simple name in a
dataset name that consists of a simple name only – the first qualifier. In clause (d),
396 Arlene Harvey
however, the variable being sought is not the name ‘the first qualifier’ (since we have
already been introduced to this name), but rather, which element is the first qualifier in a
qualified name (in this case, it is the first simple name in the qualified name). The marked
Ir ^ Id information structure in clause (d) (a case of textual deautomatization) suggests a
disruption. Two different Values have, in effect, been assigned to the same Token (the first
qualifier) (i.e. Value [A, B] = Token). The resulting lack of exhaustiveness between the
Token and Value is signaled by the logical deautomatization in both clauses (c) and (d) (in
the conditional clauses if …). Thus, although clause (d) is unusual in the combination of
decoding and specific naming/defining verb, we are alerted to its atypicality by the
deautomatization of both textual and logical features.
The decoding glossary definition in (27) below is also unusual. Although it
combines a one-to-one relation and a specific naming/defining verb, the textual
deautomatization, reflected in the marked information structure (Ir ^ Id), alerts us to the
atypical lack of equivalence. A specific class word is missing (its place is held by the
demonstrative that; see Litowitz 1977) and the definition is incomplete.14
(27) Resource: That [0 class] requiring protection by RACF (Value/Ir) is called a
resource (Tk/Id). (Decoding) (150)
(Probe: “what is that requiring protection by RACF called?” NOT “what is a
resource?”)
The encoding (defining) clause in (28) contains the verb be rather than a specific
verb and constructs a relation of non-equivalence. The term unidentified user has two
(related) meanings and this logical extension (deautomatization) is explicitly signaled by
the adverb also, and a reason for the extension is provided (because …).
(28) A user who uses the data center without specifying the user-ID (Vl/Ir) is called an
unidentified user (Tk/Id). A user who is not registered in the RACF management
dataset (Vl/Id) is also an unidentified user (Tk/Ir) || because the user does not have a
user-ID. (Encoding) (175)
The following encoding (defining/naming) clause is also incomplete: The Value
nominal group does not contain a head and represents a logical under-extension signaling
non-equivalence.
(29) The protection method for shared DASD datasets depends on how these datasets are
named. How the datasets are named (Vl/Id) [0 Head] is called the naming rule
(Tk/Ir). (Encoding) (107)
14
The marked information structure (Ir ^ Id) may have been influenced by the fact that demonstratives do not usually appear in a position of tonic prominence in the clause (i.e. in clause-final position) (Halliday and Hasan 1976).
The clause in (30) below represents the only exception to the general pattern
described in this paper. This clause is an encoding defining clause. It contains the verb be
Relational clauses in English technical discourse 397
but nonetheless expresses a relation of equivalence between the two nominal groups
(which the previous examples do not). The definition is complete and the relation between
the Token and Value is exhaustive, categorical and one-to-one. Nor is there any
deautomatization of interpersonal, textual, or logical features.
(30) User-Password Change Inhibit Interval
Certain restrictions apply to the user-passwords recorded. If a user-password is
changed many times, the same user-password may be reset. To prevent a user-
password from being constantly changed, an interval is set during which the user-
password cannot be changed. This interval (Vl/Id) is the user-password change
inhibit interval (Tk/Ir) (Encoding) (206)
This clause is fascinating since, during the initial data collection (from the manuals
and glossary), I copied it down, incorrectly as I later discovered, as:
(30’) This interval is called the user-password change inhibit interval.
Although I did not realize it at that time, my instincts about verb choice in
identifying clauses may well have been responsible for the copying error.
7. Conclusion
The most common-sense explanation for choosing specific verbs in relational clauses over
the verb be is to avoid the ambiguity associated with be. Given its potential for ambiguity,
it is somewhat surprising that be is often perceived to be the default choice, that is, the
option chosen unless there is some reason for choosing otherwise. In this paper I have
attempted to shed some light on this issue by analysing verb choice in a relatively small
corpus of identifying clauses from English technical manuals. I have argued that a semantic
complementarity exists between the general and under-specified verb be and clauses in the
decoding orientation, whose middle ergativity renders them semantically non-exhaustive.
On the other hand, specific lexical verbs (e.g. call, mean) are apparently more semantically
compatible with clauses in the encoding orientation, especially those whose double agency
allows for a human to assign the Value to the Token, and vice versa.
More generally, I have proposed that verb choice patterns in my data can be
explained by invoking the notion of equivalence. Equivalent relations are both one-to-one
and exhaustive, and relational functions (e.g. defining, naming, exemplifying, and so on)
tend to be, in the default case, either one-to-one or one-to-many (but exhibit more
variability with respect to exhaustiveness). Simply stated, equivalent relations prefer
specific verbs, while non-equivalent relations prefer be (or graphological relators),
regardless of coding orientation (but keeping in mind the typical combinations). What is
perceived to be the typical equivalence status of a particular function can be disrupted and
the incongruence between our expectations of the relationship between equivalence, coding
orientation and verb type is usually signaled by the deautomatization (foregrounding) of
non-experiential features in the clause (e.g. logical, interpersonal or textual).
Deautomatization of these meanings in general makes the relation more open and
negotiable, and less impenetrable.
398 Arlene Harvey
To test these principles further, the analysis of a larger corpus of identifying clauses
in technical discourse is necessary. A cross-comparison with other formal registers, e.g.
scientific, academic, bureaucratic, would also help explain registerial variation in the area
of knowledge construction. The kinds of patterns that occur in formal registers in
languages other than English could also be studied. One issue to keep in mind in
interpreting the results from this study is that the source manuals were drafts (not finished
products), and were, in their original form, translated from Japanese. This may account for
some of the more unusual patterns and could well have implications for teachers and
learners of English, as well as translators.
With respect to the probabilities with which various grammatical features appear in
different registers, Halliday (1991a) has noted that speakers of a language are able to make
informed guesses about the relative frequencies of particular words. This informed
guesswork represents one “aspect of knowing the language” (1991a: 35). Native speakers
or writers of English may well have an instinctive understanding of when to use be and
when to use more specific verbs based on probabilities as well as a deeply rooted (but
difficult to explain) knowledge of what is, and what is not.
References
Davidse, K. (1991) Categories of experiential grammar. Ph.D. thesis, University of Leuven.
Davidse, K. (1992a) A semiotic approach to relational clauses. Occasional Papers in Systemic Linguistics 6:
99-131.
Davidse, K. (1992b) Transitivity/ergativity: The janus-headed grammar of actions and events. In M. Davies &
L. Ravelli (eds.), Advances in systemic linguistics: Recent theory and practice. London: Pinter, pp. 103-135.
Davidse, K. (1996) Turning grammar on itself: Identifying clauses in linguistic discourse. In M. Berry, C.
Butler, R. Fawcett & G. Huang (eds.), Meaning and form: Systemic functional interpretations (Meaning and
choice in language: Studies for Michael Halliday). Ablex: Norwood, New Jersey, pp. 367-393.
Flowerdew, J. (1991) Pragmatic modifications on the ‘representative’ speech act of defining. Journal of
Pragmatics 15: 253-264.
Fries, P. (1994) On theme, rheme and discourse goals. In M. Coulthard (ed.), Advances in written text
analysis. London: Routledge, pp. 229-249.
Givón, T. (1995) Functionalism and grammar. Amsterdam: John Benjamins.
Halliday, M.A.K. (1967) Notes on transitivity and theme in English (Parts 1 and 2). Journal of Linguistics
3.1: 37-81 & 3.2: 199-244.
Halliday, M.A.K. (1968) Notes on transitivity and theme in English (Part 3). Journal of Linguistics 4.2: 179-
215.
Halliday, M.A.K. (1978) Language as social semiotic. London: Edward Arnold.
Relational clauses in English technical discourse 399
Halliday, M.A.K. (1982) The de-automatization of grammar: From Priestly’s ‘An Inspector Calls’. In J.
Anderson (ed.), Language form and linguistic variation: Papers dedicated to Angus McIntosh. Amsterdam:
Benjamins, pp. 129-59.
Halliday, M.A.K. (1988) On the ineffability of grammatical categories. In J. Benson, M. Cummings & W.
Greaves (eds.), Linguistics in a systemic perspective. Amsterdam: Benjamins, pp. 27-51.
Halliday, M.A.K. (1991a) Corpus studies and probabilistic grammar. In K. Aijer & B. Altenberg (eds.),
English corpus linguistics: Studies in honour of Jan Svartvik. London & New York: Longman, pp. 30-43.
Halliday, M.A.K. (1991b) Towards probabilistic interpretations. In E. Ventola (ed.) Functional and systemic
linguistics: Approaches and uses. Berlin, New York: Mouton de Gruyter, pp. 39-61.
Halliday, M.A.K. (1994) Introduction to functional grammar. (2nd
Ed.) London: Edward Arnold.
Halliday, M.A.K. & R. Hasan (1976) Cohesion in English. London & New York: Longman.
Halliday, M.A.K. & J. R. Martin (1993) Writing science: Literacy and discursive power. London: Falmer.
Halliday, M.A.K. & Z. James (1993) A quantitative study of polarity and primary tense in the English finite
clause. In J. Sinclair, M. Hoey & G. Fox (eds.), Techniques of description: Spoken and written discourse.
London: Routledge, pp. 32-66.
Harvey, A. (1996) Equivalence and depersonalisation in definitions: An exploration of lexicogrammatical
and rhetorical patterns in English technical discourse. Ph.D. thesis, University of Sydney.
Harvey, A. (1999) Definitions in English technical discourse: A study in metafunctional dominance and
interaction. Functions of Language 6.1: 53-94.
Jakobson, R. (1932/1984) The structure of the Russian verb. In L. Waugh & M. Halle (eds.), Russian and
Slavic grammar studies, 1931-1981. Berlin, New York & Amsterdam: Mouton, pp. 1-14.
Langacker, R. (1987) Foundations of cognitive grammar. Stanford, California: Stanford University Press.
Lemke, J. (1990a) Talking science: Language, learning and values. Norwood, N.J.: Ablex.
Lemke, J. (1990b) Technical discourse and technocratic ideology. In M.A.K. Halliday, J. Gibbons & H.
Nicholas (eds.), Learning, keeping, and using language. Amsterdam & Philadelphia: John Benjamins, pp.
435-460.
Litowitz, B. (1977) Learning to make definitions. Journal of Child Language 4: 289-304.
Martin, J.R. (1992) English text: System and structure. Amsterdam: Benjamins.
Matthiessen, C. (1995a) Lexicogrammatical cartography: English systems. Tokyo: International Language
Sciences Publishers.
Matthiessen, C. (1995b) Theme as an enabling resource in ideational ‘knowledge’ construction. In M.
Ghadessy (ed.), Thematic development in English texts. London: Pinter, pp. 20-55.
Mukarovsky, J. (1977) Word and verbal art: Selected essays. New Haven: Yale University Press.
Nesbitt, C. & G. Plum (1988) Probabilities in a systemic grammar: The clause complex in English. In R.
Fawcett & D. Young (eds.), New developments in systemic linguistics, Vol 2: Theory and application.
London: Frances Pinter, pp. 6-38.
400 Arlene Harvey
Newman, J. (in prep) (ed.) The linguistics of sitting, standing and lying.
Plum, G. and A. Cowling (1987) Social constraints on grammatical variables: Tense choice in English. In R.
Steele & T. Threadgold (eds.), Language topics: Essays in honour of Michael Halliday. Amsterdam: John
Benjamins, pp. 281-305.
Plum, G., C. Matthiessen, M. O’Donnell, L. Zeng, C. Nesbitt, A. Harvey, M.A.K. Halliday & J. Bateman
(1992) EDA research and development: Final report. Internal Document, Fujitsu Australia Ltd.
Quirk, R., S. Greenbaum, G. Leech & J. Svartvik (1985) A comprehensive grammar of the English language.
London: Longman.
Temmerman, M. (1994) Definities in schooltaal: Communicatieve, cognitieve en tekstuele aspecten van
definiëren in de derde graad van het basisonderwijs. (2 volumes). Ph.D. Thesis, Antwerpen: Universitaire
Instelling Antwerpen.
Van Oosten, J. (1984) Sitting, standing and lying in Dutch: A cognitive approach to the distribution of the
verbs zitten, staan, and liggen. In J. van Oosten & J. Snapper (eds.), Dutch linguistics in Berkeley. Berkeley:
UCB, pp. 137-160.
Wignell, P., J.R. Martin & S. Eggins (1993) The discourse of geography: Ordering and explaining the
experiential world. In M.A.K. Halliday & J.R. Martin (eds.), Writing Science: Literacy and discursive power.
London & Washington, DC: Falmer.